0.3.1

Documentation

• The README now describes the 0.3.0 robustness improvements: messy-numeric
  coercion with preserved categories, boolean-column handling, binary
  classification metrics for any pair of labels with the recorded positive
  class, class-imbalance safety through a stratified split and capped
  cross-validation folds, and clear MudraError failures that name the
  offending column. There are no code changes in 0.3.1.